计算机工程与应用 ›› 2022, Vol. 58 ›› Issue (6): 134-141.DOI: 10.3778/j.issn.1002-8331.2009-0477

• 模式识别与人工智能 • 上一篇    下一篇

基于情境信息迁移的因子分解机推荐算法

孙雨新,曹晓梅,王少辉   

  1. 南京邮电大学 计算机学院,南京 210000
  • 出版日期:2022-03-15 发布日期:2022-03-15

Factorization Machine Recommender Algorithm Based on Context Information Transfer

SUN Yuxin, CAO Xiaomei, WANG Shaohui   

  1. School of Computer, Nanjing University of Posts & Telecommunications, Nanjing 210000, China
  • Online:2022-03-15 Published:2022-03-15

摘要: 传统推荐算法大多使用用户评分数据来推测用户偏好,仅用评分数据会导致推荐结果单一,缺乏多样性和个性化,同时评分数据还普遍存在严重的稀疏性问题。针对上述问题,提出了一种基于情境信息迁移的因子分解机推荐算法。根据情境信息对数据集进行划分,利用自适应增强方法对不同情境下的数据样本进行迁移处理,将处理后的数据集放入因子分解机,实现评分预测。实验结果表明该算法能在充分使用数据样本、缓解稀疏性问题同时,进行更准确的个性化推荐,相较于传统推荐算法推荐误差降低了2.05%。

关键词: 情境信息, 迁移学习, 因子分解机, 个性化推荐

Abstract: Most of the traditional recommendation algorithms use user rating data to infer user preferences, which will lead to single recommendation results, lack of diversity and personalization, and there is a serious problem of sparse rating data. In response to the above questions, this paper proposes a factorization machine recommendation algorithm based on situation information transfer. The data set is divided by the situation information, and the adaptive enhancement method is used to migrate the data samples under different scenarios and the processed data set is put into the factorization machine to realize the score prediction. The experimental results show that the algorithm can fully use data samples to alleviate the sparsity problem. Meanwhile, it can make more accurate personalized recommendations. Compared with traditional recommendation algorithms, the recommendation error is reduced by 2.05%.

Key words: situational information, transfer learning, factorization machine, personalized recommendation